Traffic Sign Recognition Based on Improved SSD Model
Aiqing Huo, Wenle Zhang, Yi Li · 2020
In view of the problems of missed detection and low detection accuracy of the SSD model in the detection of small targets, an improved SSD model for traffic road sign recognition was proposed. The model uses DenseNet to replace SSD's basic network VGG16 to reduce the amount of network parameters; learn from the feature fusion method of FCN to improve the detection ability of small targets; increase the convolution of holes to expand the perception domain and reduce the loss of small target information; use depth Separable convolution replaces the Maxpooling layer in DenseNet to avoid information loss during feature extraction. Experiments were carried out on the CTSD data set. The experimental results show that the improved SSD model improves the recognition accuracy and recall rate.